Add dod-guard under Code Quality & Testing
@atoslinschecks…add-dod-guard → master1 files · +1 −0updated 3w agoloading diff…
@atoslinschecks…add-dod-guard → master1 files · +1 −0updated 3w ago@atoslinschecks…add-dod-guard → master1 files · +1 −0updated 3w agoAdds dod-guard under Code Quality & Testing.
DoD-Guard enforces the Definition of Done as an executable barrier for Claude Code. It targets the specific failure mode where LLM-driven coding assistants declare tasks "done" while leaving stubs, tautological tests, unverified claims, or silent regressions.
| Layer | Count | What it stops |
|---|---|---|
| Deterministic detectors (bash + python, no LLM) | 11 | stubs, TODOs, empty functions (AST), suspicious returns, tautological tests, NotImpl markers |
| Lifecycle hooks | 5 | bad edits in real time; Stop refuses to release the turn while DoD is unmet, with stop_hook_active loop prevention; --no-verify rejected |
| Adversarial subagents (read-only) | 7 | completeness, test quality, e2e proof, regressions, hostile review, claim validation, final-judge aggregator |
| Slash commands | 8 | /dod:init, /dod:verify, /dod:audit, /dod:report, /dod:stubs, /dod:tests, /dod:checklist, /dod:confess |
| Skills | 4 | reframe orchestrator behavior in any project that has .dod-guard.json |
shellcheck -x clean on every shell script.claude plugin validate passes for both plugin.json and marketplace.json.It overlaps with code-review and test-writer-fixer in spirit but operates a tier deeper: rather than catching style issues or generating tests, it audits completion claims — the most common silent failure mode of LLM-driven workflows.
Repo: https://github.com/atoslins/dod-guard Release: https://github.com/atoslins/dod-guard/releases/tag/v0.1.0